### Metric

Submissions will be evaluated using the **mean Average Precision (mAP50-95)** metric for object detection, a standard measure used to evaluate the performance of object detectors. The mAP50-95 score evaluates how well the predicted bounding boxes and class labels match the ground truth across various Intersection over Union (IoU) thresholds.

### Submission File

The submission file for this competition is a CSV formatted as follows: Each row corresponds to a single image and contains four columns: `image_id`, `prediction_string`, `image_width`, `image_height`, `image_height` and 'Usage'. The `image_id` column specifies the unique identifier for each image (image file name), while `image_width` and `image_height` provide the dimensions of the image in pixels. The `prediction_string` column consolidates all detected bounding boxes for an image into a single, space-separated string. Each detection is represented by `{label confidence x_center y_center width height}`, where `label` indicates the class ID, `confidence` represents the detection confidence score, and `x_center`, `y_center`, `width`, `height` denote the bounding box normalized coordinates in YOLO format. The 'Usage' column is a fixed value set to "Public" because this competition includes only a public leaderboard. This column exists solely to meet Kaggle's requirements.

For example:

```
002084.jpg,7 0.9342 0.3527 0.5056 0.6568 0.9747 7 0.7538 0.4666 0.5264 0.9202 0.9442,500,375,Public
007875.jpg,7 0.9515 0.4752 0.5047 0.9505 0.9732,414,500,Public

```
**The provided submission code is intended to prevent any potential conflicts, so you only need to prepare a trained YOLOv5 model.** Submit your predictions through the competition platform, and they will be automatically scored based on the ground truth annotations.